eahil 2015 - improving efficiency and confidence in systematic literature searching
TRANSCRIPT
Improving efficiency and confidence in systematic
literature searching
Wichor M. Bramer - Gerdien B. de Jonge Erasmus MC - Rotterdam the Netherlands
EAHIL+ICAHIS+ICLC – Edinburgh (UK) – June 11nd / 12th 2015
MEDLINE (OVID)1 exp Psychotherapy/ (150715)2 Problem Solving/ (21235)3 psychotherap*.mp. (68147)4 ((cogniti* or family or behavior* or behaviour* or psychological*) adj5 (intervention* or treatment* or therap*)).mp. (107111)5 (problem* adj5 solv*).mp. (54029)6 CBT.mp. (5158)7 or/1-6 (265435)8 exp Parents/ (72641)9 exp Family/ (236222)10 Caregivers/ (21502)11 (parent* or mother* or father* or family or families or caregiver* or care-giver*).mp. (1252254)12 or/8-11 (1284976)13 exp Child/ (1543748)14 exp Infant/ (934743)15 Adolescent/ (1611500)16 (child* or infant* or adolesc* or baby or babies or toddler* or teenager* or youth*).mp. (3174766)17 or/13-16 (3174766)18 exp Pain/ (306198)19 exp Complex Regional Pain Syndromes/ (4336)20 exp Rheumatic Diseases/ (175661)21 exp Neoplasms/ (2561658)22 exp Diabetes Mellitus/ (314181)23 exp Asthma/ (105870)24 exp Brain Injuries/ (49346)25 exp Inflammatory Bowel Diseases/ (60373)26 exp Anemia, Sickle Cell/ (17799)27 exp Skin Diseases/ (810983)28 exp Genital Diseases, Female/ (358725)29 exp menstruation disturbances/ (24223)30 (pain* or headache*).mp. (582940)31 (rheumat* or arthriti* or fibromyalgia).mp. (232887)32 (cancer* or neoplas* or tumor* or tumour* or malignan* or carcinoma*).mp. (2952920)33 diabet*.mp. (471099)34 asthma*.mp. (140539)35 (brain adj5 (trauma* or injur*)).mp. (68497)36 (bowel* adj5 inflammatory adj5 (condition* or disease* or illness*)).mp. (31393)37 (sickle cell adj5 (disease* or disorder* or anemia*)).mp. (19955)38 ((skin adj5 (disease* or disorder*)) or eczema*).mp. (105706)39 ((gynecologic* or gynaecologic*) adj5 (disease* or disorder*)).mp. (5081)40 dysmenorrh*.mp. (5152)41 endometriosis.mp. (20182)42 Chronic Disease/ (219791)43 ((chronic* or longterm or long-term) adj5 (condition* or ill* or disease*)).mp. (418976)44 or/18-43 (5469388)45 randomized controlled trial.pt. (376608)46 controlled clinical trial.pt. (88576)47 randomized.ab. (297316)48 placebo.ab. (155203)49 drug therapy.fs. (1708731)50 randomly.ab. (214956)51 trial.ab. (308738)52 groups.ab. (1366808)53 45 or 46 or 47 or 48 or 49 or 50 or 51 or 52 (3363144)54 exp animals/ not humans.sh. (3954113)55 53 not 54 (2885271)56 7 and 12 and 17 and 44 and 55 (1119)57 (201203* or 201204* or 201205* or 201206* or 201207* or 201208* or 201209* or 201210* or 201011* or 201212* or 2013* or 2014*).ed. (2350810)58 56 and 57 (193)
EMBASE (OVID)1 exp Psychotherapy/ (188395)2 Problem Solving/ (26195)3 psychotherap*.mp. (96298)4 ((cogniti* or family or behavior* or behaviour* or psychological*) adj5 (intervention* or treatment* or therap*)).mp. (213343)5 (problem* adj5 solv*).mp. (62494)6 CBT.mp. (7776)7 or/1-6 (385750)8 exp Parents/ (155722)9 exp Family/ (335975)10 Caregivers/ (39951)11 (parent* or mother* or father* or family or families or caregiver* or care-giver*).mp. (1393196)12 or/8-11 (1438650)13 exp Child/ (2083396)14 exp Infant/ (860556)15 Adolescent/ (1219788)16 (child* or infant* or adolesc* or baby or babies or toddler* or teenager* or youth*).mp. (2974509)17 or/13-16 (3192664)18 exp Pain/ (829233)19 exp Complex Regional Pain Syndromes/ (7323)20 exp Rheumatic Diseases/ (188325)21 exp Neoplasms/ (3394555)22 exp Diabetes Mellitus/ (598732)23 exp Asthma/ (188864)24 exp Brain Injuries/ (122563)25 exp Inflammatory Bowel Diseases/ (3120)26 exp Anemia, Sickle Cell/ (26998)27 exp Skin Diseases/ (1197204)28 exp Genital Diseases, Female/ (490160)29 exp menstruation disturbances/ (51427)30 (pain* or headache*).mp. (971422)31 (rheumat* or arthriti* or fibromyalgia).mp. (348433)32 (cancer* or neoplas* or tumor* or tumour* or malignan* or carcinoma*).mp. (3603831)33 diabet*.mp. (709575)34 asthma*.mp. (213871)35 (brain adj5 (trauma* or injur*)).mp. (124711)36 (bowel* adj5 inflammatory adj5 (condition* or disease* or illness*)).mp. (43873)37 (sickle cell adj5 (disease* or disorder* or anemia*)).mp. (26819)38 ((skin adj5 (disease* or disorder*)) or eczema*).mp. (145349)39 ((gynecologic* or gynaecologic*) adj5 (disease* or disorder*)).mp. (18632)40 dysmenorrh*.mp. (9538)41 endometriosis.mp. (27781)42 Chronic Disease/ (149689)43 ((chronic* or longterm or long-term) adj5 (condition* or ill* or disease*)).mp. (512959)44 or/18-43 (7140484)45 random$.tw. (892218)46 factorial$.tw. (23405)47 crossover$.tw. (49675)48 cross over$.tw. (22262)49 cross-over$.tw. (22262)50 placebo$.tw. (203743)51 (doubl$ adj blind$).tw. (147609)52 (singl$ adj blind$).tw. (14595)53 assign$.tw. (241607)54 allocat$.tw. (84725)55 volunteer$.tw. (181433)56 Crossover Procedure/ (39305)57 double-blind procedure.tw. (219)58 Randomized Controlled Trial/ (347009)59 Single Blind Procedure/ (18444)60 or/45-59 (1428276)61 (animal/ or nonhuman/) not human/ (4623513)62 60 not 61 (1262127)63 7 and 12 and 17 and 44 and 62 (952)64 (201203* or 201204* or 201205* or 201206* or 201207* or 201208* or 201209* or 201210* or 201011* or 201212* or 2013* or 2014*).dd. (3247786)65 63 and 64 (263)66 limit 65 to embase (204)
0 5 10 15 20 25 30 35 40 45 500%
25%
50%
75%
100%
Meta-analysis (N=148)
Hours needed to create search strategy
Perc
enta
ge o
f Sys
tem
atic
Revi
ews
Hausner, 2015 (13); Saleh, 2014 (17); Lyon, 2014 (9); Gann, 2013 (47); Allen, 1999 (19) [1-5] and unpublished data from London Health Sciences Centre in Canada (43)
0 5 10 15 20 25 30 35 40 45 500%
25%
50%
75%
100%
Meta-analysis (N=148)
Bramer (N=258)
Hours needed to create search strategy
Perc
enta
ge o
f Sys
tem
atic
Revi
ews
Hausner, 2015 (13); Saleh, 2014 (17); Lyon, 2014 (9); Gann, 2013 (47); Allen, 1999 (19) [1-5] and unpublished data from London Health Sciences Centre in Canada (43)
How well do they perform?
Benchmark This study# terms 46 54# databases 3 7# hits 1399 2224# includes 19 25precision 1.2% 2.0%Time needed 960 min 70 min
Benckmark: 93 systematic reviews from 2014 with an acknowledgement or co-authorship of a medical information specialist from another Dutch academic hospital.
Systematic searching
1. Examining the research question (practicum)2. Identifying key terms (practicum)
a. Thesaurus termsb. Free text terms
3. Database / interface choice (discussion)4. Creating a basic search strategy (practicum)5. Optimizing the basic strategy (practicum)6. Translating the strategy between databases
(demonstration) 7. Increasing specificity or sensitivity (demonstration)8. Filters (discussion)9. Finding errors (demonstration)
Which articles can answer the question?
Facilitators and barriers in preconception care
evaluation studies of preconception care
Predictive factors for oronasal fistulas in children treated for clef lip
follow up studies of children treated for clef lip Study types not as exclusive filters
but additionally added in elements
Elements : Bias
Very specific characteristicsDuration of breastfeeding and intelligence
JudgementsPoor family functioning, harsh parenting and bullying behavior
Why a bias?
Elements : Duplicates
Pre-eclampsia AND pregnant AND women
Lichtenstein therapy AND inguinal hernias
Which elements to use:Does eating Broccoli help in the prevention of cancer
generalspecific
important
unimportant
Broccoli Cancer
Prevention
Broccoli Cancer Prevention
Which elements to use:your own research question
generalspecific
important
unimportant
PRACTICUM Describe the articles you want to find
Determine elements Plot them in the graph
Discuss with your neighbour
Identifying terms
Two major thesauri:MeSH (Medline)Emtree (Embase)
Three interfaces:PubMed (MeSH)Ovid (MeSH and Emtree)Embase.com (Emtree)
Identifying thesaurus terms
One element Multiple thesaurus terms (OR)
Multiple elements One thesaurus term
One element Multiple thesaurus terms (AND)
No thesaurus terms found
Truncate the search terms with * when searching MeSH database in PubMed
tear vs. tear*• Simple search in PubReminer• Use [tt] in PubMed to find translations• Use Google, Wikipedia etc. to find alternative terms
If all this fails:Use only free text words and see what you get
Identifying free text words
Optimal sensitivity only by additionally searching in title / abstract:• Relevant, non-inverted terms from synonyms /
entry terms• Relevant words / phrases from narrower terms
Add for yourself
- Abbreviations- Also partial abbreviations: Hip OA / Sentinel LN
- Alternative spelling- Randomization / Randomisation - Pediatric / Paediatric - Tumor / Tumour
PRACTICUM (5 min) – Finding keywords
Using the MeSH interface from PubMedTry to find as much search terms as possible for
Kidney dialysis
Renal Dialysis [MeSH]Renal Dialy* Hemodialy* Extracorporeal Dialy*Hemodiafiltration* Peritoneal Dialy*Kidneys, Artificial [mesh]Artificial Kidney*Blood Dialy*Acetate Free Biofiltrat* CAPDRenal Replacement Therapy[mesh:noexp]Renal Replace* Kidney Replace*
Haemodialy*Haemodiafiltrat*
(Dialysis [mesh] AND (Kidney Diseases [mesh] OR Kidney [mesh]))
Kidney dialy*
Database / interface choice
Advice for start database
1. embase.com if available2. Otherwise embase with proximity
(preferably Ovid)3. No embase? Medline with proximity4. No proximity interface available? PubMed
– more complicated – will miss relevant articles – or get much more noise
Creating a basic search strategy
Embase.com Ovid EBSCOHost PubMedBoolean AND, OR, NOTThesaurus '…'/exp exp "…"/ DE "…+" "…"[mh]Title / abstract
():ab,ti ().ab,ti. TI() OR AB() [tiab]
Proximity (() NEAR/n ()) (() ADJn ()) (() Nn ()) (() AND ()) truncated phrase*
Standard syntaxExtended version (20 syntaxes in 7 databases) as supplement in handouts
DEMO
Creating a basic search strategy
General rules:1. Work in a word document, not the interface2. Create single line search strategies
Single line approach or set numbers?Well structured?#1 brassica/exp#2 brassica*:ab,ti#3 broccoli:ab,ti#4 #1 OR #2 OR #3#5 neoplasm/exp#6 neoplas*:ab,ti#7 cancer*:ab,ti#8 tumor*:ab,ti#9 #5 OR #6 OR #7 OR #8#10 #4 AND #9#11 [animals]/lim#12 [humans]/lim#13 #11 NOT #12#14 #10 NOT #13
Unstructured?
(brassica/exp OR (brassica* OR broccoli):ab,ti) AND (neoplasm/exp OR (neoplas* OR cancer* OR tumor*):ab,ti) NOT ([animals]/lim NOT [humans]/lim)
Creating a basic search strategy
General rules:1. Work in a word document, not the interface2. Create single line search strategies3. Type code before adding the words4. After opening parentheses, immediately close them5. Never remove parentheses, only add6. Prepare proximity statement for reuse7. Copy terms from thesaurus instead of typing them8. Use arrow keys and Ctrl-shortcuts9. Use ab,ti to create the strategy, final syntax can be broad
PRACTICUM
Optimizing search strategy
Search temporarily per element: (thesaurus) NOT (title/abstract)(title/abstract) NOT (thesaurus)
Remove all field codes (automatic term mapping)
Check retrieved relevant articles for missed terms
DEMO + PRACTICUM
Translating the strategy between databases
Embase.com (Embase and Medline)Thorough optimization of thesaurus terms and title / abstract terms in search strategy
Macro and edit thesaurus
Medline (Ovid)MeSH terms in PubMed / Optimization
Add extra terms Remove thesaurus
Cochrane library
Macro
Web-of-science
Macro
Scopus
Macro
PubMed publisher
[sb]
Macro and edit
thesaurus
CINAHL (EBSCO)
edit thesaurus
PsycINFO (OvidSP)
Macro and edit
Google Scholar
Macro and edit
Lilacs / Scielo / ProQuest / Clinical trials etc
Translation
Ctrl-H: find and replacespecial tip: ^p for linebreak^t for tab
Macros: bit.ly/emcmacros
Considerations
Explode a broad term or not?Start exploded, when many irrelevant results appear,
be more specificTruncate a term or not?
Truncate as much as possible, hardly noisePhrases or proximity?
Always proximity (when available in interface), easier and more sensitiveHow broad should proximity be?
Start with 3, increase with 3 andevaluate newly retrieved articles
• Remove too general terms or too broad thesaurus terms.
• Do not explode thesaurus terms• Add an extra element• (mesh) AND (subheadings)
(mesh/subheading)• Narrow proximity• Separate elements proximity• Remove truncation (longer wordstem)• (I OR C) (I) AND (C)• Add Filters for study type?
date? language?
Only if you can afford missing relevant refs:• For the most important element: use major
and title instead of keyword and title abstract
Increase Sensitivity (always)
• Drop less important elements• Add extra words (thesaurus NOT tiab)• Generalize specific elements (broader
thesaurus terms)• Explode thesaurus terms• (mesh/subheading) (mesh) AND
(subheading)• Broaden proximity • Proximity separate words (AND)• Add truncation (shorten wordstem)• (I) AND (C) (I OR C)
• Temporarily broaden or drop a very specific element to optimize the others
• Save relevant hits on clipboard to check sensitivity of more specific searches
Increase Specificity (if too many)
Filters
Add filters only if they are important elements in your research question
(top right corner of graph)
Where to find good filters to use?Translating between databases?
Check for errorsHow do you know an error occurred?
Sometimes: More or less hits than expected
Too often: NOT!
Check your query for mistakes
Use Ctrl-F to search for ' AND '
Scan or search for Automatic Term Mapping (such as .mp. OR [all fields])
Successful search?
1. Coverage of accessible databases2. Features of search engine of used interfaces3. Quality of indexing4. Presence and quality of abstract
5. Quality of search and experience of searcher
Success factors / Time savers
• Optimization techniques • Macros for syntax translation • Direct feedback by researcher • Database order and interface choice • Single-line search strategies• Create query in Word and paste in database• Experience (be pro-active) • Having two screens on your computer• Computer literacy
keyboard short cuts (supplement 3)
Quality is the norm, not speed!
Not a blueprint for speed but for confidence by standardization
Speed can only be a result of experience
By gaining speed more customers can be helped and overall review quality will improve
Contact details
E-mail: [email protected]@erasmusmc.nl [email protected]
Website: www.erasmusmc.nl/medbib
Twitter: @wichorLinkedIn: nl.linkedin.com/in/wichorSlideshare: www.slideshare.net/wichor
Further reading:• Bramer WM, Giustini D, Kramer BM, Anderson PF.
The comparative recall of Google Scholar versus PubMed in identical searches for biomedical systematic reviews: a review of searches used in systematic reviews. Syst Rev. 2013 Dec 23;2:115.
• Bramer WM, Holland L, Mollema J, Hannon T, Bekhuis T. Removing duplicates in retrieval sets from electronic databases: comparing the efficiency and accuracy of the Bramer-method with other methods and software packages (Eahil conference, Rome)
• Embase webinar : Systematic searching with Wichor Bramer (25 feb 2015)• Bramer WM, Pols DHJ, Bindels PJE, Bohnen AM. A fast, reliable and
objective method for creating thematic search filters in bibliographic databases [in process]
• Work in progress on coverage, retrieval and relevance ranking of the major databases for systematic reviews (>450 review searches; >100 finished; >4500 included references)
• Two articles in progress about this methodology (how to, and how well)
What to do with subheadingsSeparate element(broccoli/de OR (broccoli* OR bassica*):ab,ti) AND (neoplasm/exp OR (cancer* OR neoplas*):ab,ti) AND (prevention:lnk OR prevention/exp OR (prevent*):ab,ti)
Combined element(broccoli/de OR (broccoli* OR bassica*):ab,ti) AND (neoplasms/exp/dm_pc OR ‘cancer prevention’/de OR ((cancer OR neoplasm) NEAR/3 (prevent*)):ab,ti)
Sensitive More hits : 345 (101 unique) Less chance of missing relevant articles Recommended for start For SRs highly recommended
Specific Less hits : 257 (13 unique) Higher chance of missing relevant
articles To reduce the number of hits Use proximity in free text
Why start in embase?Source: Unpublished research Bramer (5048 included references from 111 reviews)
Relevance ranking
Which databases to useDatabase Used in % of our SRs
Embase.com 100%
Medline (Ovid) 99%
(PubMed 1%)
Cochrane central 98%
Web-of-Science 95%
Google Scholar 92%
PubMed (publisher[sb]) 91%
Scopus 36%
Cinahl 25%
PsycINFO 19%
Others (proquest, lilacs, scielo, sportdiscus, eric, amed etc)
9%
Unique articles per database
embase 509Web-of-Science 244medline 195Google Scholar 139Other databases 92
Source: own observation on 5048 included references of 111 systematic reviews